* [Xenomai] ARM L2C-310 cahce controller enabling
@ 2017-01-31 8:17 Mauro Salvini
2017-01-31 8:28 ` Mauro Salvini
0 siblings, 1 reply; 4+ messages in thread
From: Mauro Salvini @ 2017-01-31 8:17 UTC (permalink / raw)
To: xenomai
Hi,
I'm working on a iMX6SX custom platform and I'm trying to use Xenomai on
top of it.
In the past, I tried Xenomai on a iMX6SX SabreSD demoboard using kernel
branch 4.1.15 from Freescale patched with Xenomai 3.0.1 (some little
rejections corrected by hand), and I was able to get a maximum latency
around 25us.
Now, using kernel 4.1.36 from Freescale Community (that integrates the
Freescale kernel branch) patched with Xenomai 3.0.3 I get maximum
latencies around 45us.
Trying to figure out what is the source(s) of increased latencies, I
found these two kernel logs:
[ 0.000000] L2C: I-pipe: revision >= L310-r3p2 detected, forcing WA.
[ 0.000000] L2C: I-pipe: write-allocate enabled, induces high
latencies.
I wonder why L2C-310 cache controllers that are greater or equal to r3p2
revision must be forced enabled.
In past I read (and asked) in this mailing list that L2 cache should be
disabled to avoid higher latencies (as kernel log says).
Thanks in advance, best regards
Mauro
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Xenomai] ARM L2C-310 cahce controller enabling
2017-01-31 8:17 [Xenomai] ARM L2C-310 cahce controller enabling Mauro Salvini
@ 2017-01-31 8:28 ` Mauro Salvini
2017-01-31 9:34 ` Philippe Gerum
0 siblings, 1 reply; 4+ messages in thread
From: Mauro Salvini @ 2017-01-31 8:28 UTC (permalink / raw)
To: xenomai
Hi,
sorry, found myself the answer here:
http://git.xenomai.org/ipipe.git/commit/?h=ipipe-4.1.y&id=a2a9a6148449095fa658a6d0d0cbed3308be2cfb
So, should be L2 cache disabled for iMX6SX (that has a single core)?
Thanks again, regards
Mauro
On Tue, 2017-01-31 at 09:17 +0100, Mauro Salvini wrote:
> Hi,
> I'm working on a iMX6SX custom platform and I'm trying to use Xenomai on
> top of it.
> In the past, I tried Xenomai on a iMX6SX SabreSD demoboard using kernel
> branch 4.1.15 from Freescale patched with Xenomai 3.0.1 (some little
> rejections corrected by hand), and I was able to get a maximum latency
> around 25us.
> Now, using kernel 4.1.36 from Freescale Community (that integrates the
> Freescale kernel branch) patched with Xenomai 3.0.3 I get maximum
> latencies around 45us.
>
> Trying to figure out what is the source(s) of increased latencies, I
> found these two kernel logs:
>
> [ 0.000000] L2C: I-pipe: revision >= L310-r3p2 detected, forcing WA.
> [ 0.000000] L2C: I-pipe: write-allocate enabled, induces high
> latencies.
>
> I wonder why L2C-310 cache controllers that are greater or equal to r3p2
> revision must be forced enabled.
>
> In past I read (and asked) in this mailing list that L2 cache should be
> disabled to avoid higher latencies (as kernel log says).
>
> Thanks in advance, best regards
>
> Mauro
>
>
> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> https://xenomai.org/mailman/listinfo/xenomai
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Xenomai] ARM L2C-310 cahce controller enabling
2017-01-31 8:28 ` Mauro Salvini
@ 2017-01-31 9:34 ` Philippe Gerum
2017-02-01 8:06 ` Mauro Salvini
0 siblings, 1 reply; 4+ messages in thread
From: Philippe Gerum @ 2017-01-31 9:34 UTC (permalink / raw)
To: Mauro Salvini, xenomai
On 01/31/2017 09:28 AM, Mauro Salvini wrote:
> Hi,
> sorry, found myself the answer here:
>
> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-4.1.y&id=a2a9a6148449095fa658a6d0d0cbed3308be2cfb
>
> So, should be L2 cache disabled for iMX6SX (that has a single core)?
Not the entire cache, but you may want to try running with
write-allocate off using the "l2x0_write_allocate" kernel switch, since
cache coherence issues are not going to be relevant in this case.
>
> Thanks again, regards
> Mauro
>
> On Tue, 2017-01-31 at 09:17 +0100, Mauro Salvini wrote:
>> Hi,
>> I'm working on a iMX6SX custom platform and I'm trying to use Xenomai on
>> top of it.
>> In the past, I tried Xenomai on a iMX6SX SabreSD demoboard using kernel
>> branch 4.1.15 from Freescale patched with Xenomai 3.0.1 (some little
>> rejections corrected by hand), and I was able to get a maximum latency
>> around 25us.
>> Now, using kernel 4.1.36 from Freescale Community (that integrates the
>> Freescale kernel branch) patched with Xenomai 3.0.3 I get maximum
>> latencies around 45us.
>>
>> Trying to figure out what is the source(s) of increased latencies, I
>> found these two kernel logs:
>>
>> [ 0.000000] L2C: I-pipe: revision >= L310-r3p2 detected, forcing WA.
>> [ 0.000000] L2C: I-pipe: write-allocate enabled, induces high
>> latencies.
>>
>> I wonder why L2C-310 cache controllers that are greater or equal to r3p2
>> revision must be forced enabled.
>>
>> In past I read (and asked) in this mailing list that L2 cache should be
>> disabled to avoid higher latencies (as kernel log says).
>>
>> Thanks in advance, best regards
>>
>> Mauro
>>
>>
>> _______________________________________________
>> Xenomai mailing list
>> Xenomai@xenomai.org
>> https://xenomai.org/mailman/listinfo/xenomai
>
>
> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> https://xenomai.org/mailman/listinfo/xenomai
>
--
Philippe.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Xenomai] ARM L2C-310 cahce controller enabling
2017-01-31 9:34 ` Philippe Gerum
@ 2017-02-01 8:06 ` Mauro Salvini
0 siblings, 0 replies; 4+ messages in thread
From: Mauro Salvini @ 2017-02-01 8:06 UTC (permalink / raw)
To: Philippe Gerum; +Cc: xenomai
Thank you Philippe.
I will try this, now I'm facing with other problems (I will send another
mail for those).
Only one note: for systems with L310 cache controller with version
greater or equal to R3P2 (like iMX6SX) the "l2x0_write_allocate" kernel
switch is useless (it's always overwritten in __l2c_init() call), will
need a code edit to disable write-allocate.
Regards
Mauro
On Tue, 2017-01-31 at 10:34 +0100, Philippe Gerum wrote:
> On 01/31/2017 09:28 AM, Mauro Salvini wrote:
> > Hi,
> > sorry, found myself the answer here:
> >
> > http://git.xenomai.org/ipipe.git/commit/?h=ipipe-4.1.y&id=a2a9a6148449095fa658a6d0d0cbed3308be2cfb
> >
> > So, should be L2 cache disabled for iMX6SX (that has a single core)?
>
> Not the entire cache, but you may want to try running with
> write-allocate off using the "l2x0_write_allocate" kernel switch, since
> cache coherence issues are not going to be relevant in this case.
>
> >
> > Thanks again, regards
> > Mauro
> >
> > On Tue, 2017-01-31 at 09:17 +0100, Mauro Salvini wrote:
> >> Hi,
> >> I'm working on a iMX6SX custom platform and I'm trying to use Xenomai on
> >> top of it.
> >> In the past, I tried Xenomai on a iMX6SX SabreSD demoboard using kernel
> >> branch 4.1.15 from Freescale patched with Xenomai 3.0.1 (some little
> >> rejections corrected by hand), and I was able to get a maximum latency
> >> around 25us.
> >> Now, using kernel 4.1.36 from Freescale Community (that integrates the
> >> Freescale kernel branch) patched with Xenomai 3.0.3 I get maximum
> >> latencies around 45us.
> >>
> >> Trying to figure out what is the source(s) of increased latencies, I
> >> found these two kernel logs:
> >>
> >> [ 0.000000] L2C: I-pipe: revision >= L310-r3p2 detected, forcing WA.
> >> [ 0.000000] L2C: I-pipe: write-allocate enabled, induces high
> >> latencies.
> >>
> >> I wonder why L2C-310 cache controllers that are greater or equal to r3p2
> >> revision must be forced enabled.
> >>
> >> In past I read (and asked) in this mailing list that L2 cache should be
> >> disabled to avoid higher latencies (as kernel log says).
> >>
> >> Thanks in advance, best regards
> >>
> >> Mauro
> >>
> >>
> >> _______________________________________________
> >> Xenomai mailing list
> >> Xenomai@xenomai.org
> >> https://xenomai.org/mailman/listinfo/xenomai
> >
> >
> > _______________________________________________
> > Xenomai mailing list
> > Xenomai@xenomai.org
> > https://xenomai.org/mailman/listinfo/xenomai
> >
>
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2017-02-01 8:06 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-01-31 8:17 [Xenomai] ARM L2C-310 cahce controller enabling Mauro Salvini
2017-01-31 8:28 ` Mauro Salvini
2017-01-31 9:34 ` Philippe Gerum
2017-02-01 8:06 ` Mauro Salvini
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.